Automated discrimination between atrial fibrillation and atrial flutter in the resting 12-lead electrocardiogram

Abstract Computerized time-domain analysis of the QRST-subtracted 12-lead electrocardiogram (ECG) has been used successfully to determine several atrial activity patterns. These time-domain methods are particularly useful for low-frequency signals such as those originating at the sinus node. However, high frequency atrial fibrillation (AFIB) and atrial flutter (AFL) waves can be better estimated by using spectral methods. In this study, we investigated the use of spectral entropy (SE) and spectral peak detection to distinguish fibrillatory from flutter activity in the QRST-subtracted ECG. In a set of 4,172 cardiologist-overread EGGs, a computerized ECG analysis program (12SL MAC-Rhythm, GE-Marquette Medical Systems, Milwaukee, WI) detected 270 AFIB rhythms and 100 AFL rhythms. Compared to the cardiologist's reading the AFIB versus AFL miss-classification error was 5.6%. The Fourier Transform was used to estimate the power spectral density of the QRST-subtracted ECG data. Individual lead spectra were then averaged and SE was computed for each of the ECGs originally called AFIB or AFL by the computer program. Additional criteria that included SE, spectral peak frequencies, and time-domain measures of atrial activity were then applied to discriminate between the 2 rhythms. Employing these criteria resulted in a decrease of miss-classification error to 2.5%.